Cellular scaling laws in the mammalian brain.

Whole-brain cell atlases show that neuronal cell type diversity peaks not in neuron-rich cortex and cerebellum but in deep subcortical structures such as the hypothalamus and brainstem. A proposed framework divides the vertebrate brain into a learning subsystem that scales by replicating repetitive modules, like adding parameters in machine learning, and a steering subsystem that scales by diversifying bespoke cell types to encode innate drives as a biological reward function. The framework offers a unified account of brain architecture, brain disease etiology, and possible artificial intelligence model design.